Why AI Is Still Unreliable ...

(and will probably remain like that)

Many today fear losing their jobs and businesses because of the rise of artificial intelligence. This is, of course, a justified fear that many people have. A machine that can generate long pieces of text in minutes, that can create incredible images and videos, that can generate code and translate discourses in over 150 languages.

Who hasn't thought of the possibility of AI replacing many of us, i.e., taking our jobs?

But there is a real-life example that shows why AI is unreliable and why humans will always be needed, even with the dawn of such interesting technology.

When AI Decieves Hundreds of People

A couple of days ago, hundreds of people landed in the city of Jurilovca, in Southeast Romania. There, many believe, was an annual fish festival, where people could listen to good music, drink wine, and, of course, eat fish.

However, things turn out completely wrong. There was no festival in this city, and nobody had ever planned for one in this city. The mayor did not know about it, not even the town's drunk.

So how did we end in such a situation?

The Failure of AI

The people who arrived in the city of Jurilovca were taken there by various travelling agencies. These agencies got the information that a fish festival would be held in the city from ChatGPT.

Yes, because ChatGPT told them that there will be a fish festival in that city, and many of these professionals in the tourism industry organized trips there without even verifying whether the information is correct or not. In other words, they entrusted ChatGPT, an LLM.

The Moral of the Story

ChatGPT is only a tool that can help you speed up certain tasks and be a helpful companion in research. However, it can never replace a human. Just like some people were fooled about a so-called fish festival in a small Romanian city, others might

- get pieces of code full of errors - receive bad translations from one language to another - craft articles full of misinformation and nonsense - make plans that don't rise to the expectations

No matter how you put it, it is never a good idea to put your trust in AI blindly and completely.

Your client should never be the first person to test the work.

A recent update to my Contra portfolio caught this perfectly. The editor preview showed the new cover. The public page still showed the old one.

It looked done until we opened the page a visitor would see.

That gap is easy to miss when AI helps you move fast. The file exists. The tool says success. The reply says 'done'. So you move on.

Before I hand over the result, I check the other end:

Open the exported file.

Click through the live page.

Try the shared link without owner access.

Check that the form submission actually arrives.

Pick the checks that match the work. A screenshot of the editor can't prove the public page changed.

AI can help make the work. Your name is still on the handoff.

What's one check you always run before sending work out?

This is such a useful reminder that quality includes the handoff, not just the artifact. Testing the live page, exported file, permissions, and real form flow is a strong practical checklist for catching the gaps screenshots can hide.